Analysis of the algorithm: From kernels to backup genes.

Kernelization section

The algorithm transformed the semantic similarity matrix to make it compatible with a kernel. Once this was done for each network and kernel type, it was integrated by kernel type. Below there is a general analysis of the properties of each matrix in the different phases of the process.

Annotations properties

Table 1. Annotation files descriptors

Net Min Max Average Standard_Deviation
biological_process 1 17 4.25 5.339653992713831
cellular_component 1 8 3.241758241758242 3.7898070350134403
molecular_function 1 7 2.3820224719101124 2.7314234672242548
phenotype 1 251 34.960396039603964 53.4410936924901

Matrix properties

Table 2. Similarity matrixes

Net Matrix_Dimensions Matrix_Elements Matrix_Elements_Non_Zero
biological_process_sim 84x84 7056 6046
cellular_component_sim 91x91 8281 8190
molecular_function_sim 89x89 7921 7832
phenotype_sim 101x101 10201 10100

Table 3. Filtered similarity matrixes

Table 4. Uncombined kernel matrixes

Net Kernel Matrix_Dimensions Matrix_Elements Matrix_Elements_Non_Zero
biological_process ct 84x84 7056 7056
biological_process el 84x84 7056 7056
biological_process ka 84x84 7056 6130
cellular_component ct 91x91 8281 8281
cellular_component el 91x91 8281 8281
cellular_component ka 91x91 8281 8281
molecular_function ct 89x89 7921 7921
molecular_function el 89x89 7921 7921
molecular_function ka 89x89 7921 7921
phenotype ct 101x101 10201 10201
phenotype el 101x101 10201 10201
phenotype ka 101x101 10201 10201

Table 5. Integrated kernel matrixes

Integration Kernel Matrix_Dimensions Matrix_Elements Matrix_Elements_Non_Zero
integration_mean_by_presence ct 200x200 40000 20186
integration_mean_by_presence el 200x200 40000 20186
integration_mean_by_presence ka 200x200 40000 20124

Weight values

Comparing types of kernel

Comparing integrations and kernel types